uBlock Origin patched a website-driven element-picker flaw that could let a hostile webpage save trusted filters. Those filters could run script in another site's context. For developers, the useful part of the story is the authority boundary: an interface owned by an extension can be reachable from a page while still needing to reject that page's commands. The maintainer's repair history shows why protecting an address was an incomplete answer.

TL;DR

  • The maintainer advisory lists 1.75.2 and lower as affected and 1.75.4 as patched. GitHub and Mozilla currently list stable 1.76.4. Use normal extension controls to update and check what is installed.
  • The documented path concerns Firefox MV2, a relevant resource-serving filtering rule, and a user click. Keep those conditions attached to the claim.
  • The first request-method hardening was bypassable. The next repair required a verified secret token to launch the picker.
  • The researcher explicitly says he did not build the hidden click overlay. His report and the advisory provide no victim count or evidence of active exploitation.

What the uBlock Origin element picker can authorize

The element picker lets a user choose an unwanted page element and create a blocking rule. Its convenience comes from keeping the interaction near the content the user wants to change. According to Smitty van Bodegom's report, its interface is an extension resource embedded over the webpage, and that interface can communicate with the extension background process to add filters.

That arrangement puts two different kinds of authority next to each other. The website supplies the content being inspected. The extension owns the tool that changes blocking behavior. A viewer can see one page-sized experience while the software needs to preserve a much stricter distinction between the parties inside it.

The report says rules added through the picker were treated as trusted, at least until recompilation. Some trusted filters can inject JavaScript into a matching page. Saving such a rule is therefore a consequential action. It can influence what happens when a different site is opened, beyond the element the user appeared to select.

I would review a feature like this by tracing the authority attached to the saved state. Who owns the interface? Which component accepts its messages? What happens to the rule after it reaches storage? Those questions expose the consequence that a tidy user interface can hide. In this episode's conceptual diagram, the page, picker and filter store stay separate so the transfer of authority remains visible.

How the reported path crossed a website boundary

A filtering feature could return an extension resource as a substitute for an unwanted network response. Van Bodegom describes this as supplying a harmless replacement for a script the website expects. The goal was compatibility while blocking unwanted content.

The same resource path could expose the extension's random internal identifier. Once a hostile page knew the internal address, it could load the picker interface and send it messages. The report describes an insufficiently checked input that could add another rule beyond the ordinary choice presented by the interface.

That second rule could reach the trusted filter store. It could target another website and execute script in that site's context. This is the boundary crossing behind the report's universal cross-site scripting terminology. The extension had authority to affect matching sites, and the hostile page could borrow that authority through the picker.

This explanation deliberately stays at the level of ownership, messages and saved state. We have not run the demonstration, installed an old extension, created a malicious filter or tested a real account. The video uses conceptual objects and a genuine recording of the public advisory's non-operational header. Neither is a claim that our show reproduced the exploit.

Firefox prerequisites and the click evidence

The advisory's summary identifies the Firefox MV2 extension, a relevant scriptlet filter that supplies an extension resource, and a user click. A vulnerability headline becomes less useful when those prerequisites disappear. Someone checking their own browser needs the correct product, architecture and installed version.

Van Bodegom also mentions a similar but less problematic issue in uBlock Origin Lite. That is a separate version and architecture. The statement supports acknowledging a related issue; it supplies no blanket conclusion that every MV3 extension is safe, or that every browser has the same reported path.

The user click is especially important to the evidence. The researcher describes making the embedded picker invisible or covering it so an ordinary click activates its Create button. He explicitly says he did not implement that hidden-overlay portion. The demonstrated ability to control the picker and the proposed way of disguising the click are distinct parts of his account.

The sources provide no victim count or evidence of active exploitation. That absence limits what we can say from this material. It cannot establish that no attack has ever occurred. I would preserve the qualification in summaries, captions and translated metadata rather than replacing it with a confident claim about all users.

Why recompilation was a weak cleanup boundary

The researcher says recompiling filters could clear the added rule's trusted status. That gives the story a tempting escape route: perhaps a later edit repairs the state before it matters. He also points out that the attacking page could immediately open the target page.

Those events have an order. The rule can be saved, the target page can be opened, and a later change can cause recompilation. A cleanup event occurring after the consequence has started offers a poor foundation for the authorization guarantee. Our diagram shows that ordering without assigning invented milliseconds or claiming a race measurement.

For an extension feature that saves privileged state, I would want the permission decision before the state change. A subsequent recompile can still have useful behavior, but its existence does not explain why an untrusted caller was permitted to create the rule in the first place. The repair history returns to exactly that question.

Why the first fix left the picker open

Hill's advisory records the private report on November 1 and a first hardening change published in 1.75.2 on November 2. Its title is "Avoid use of potentially tainted XHR methods in scriptlets." The email exchange then records a bypass that could still expose the internal identifier through another part of the request machinery.

Closing one route to a hidden address left a powerful interface reachable through another route. More importantly, the picker still needed a reliable way to distinguish its legitimate extension flow from a hostile page. The second repair addressed that launch boundary.

The November 3 change, released in 1.75.4, is titled "Launch element picker only with a verified secret token." The legitimate flow supplies authorization, and the picker checks it. A page knowing the resource address alone should remain outside the permission boundary.

Later changes in 1.76.0 validated proposed filters and restricted resource-related filters. They add protection around what gets saved and which resources filters can use. The selected advisory still names 1.75.4 as patched. The newer 1.76.4 release describes additional reported security work, and the official Mozilla listing currently offers that stable version. These statements have different scopes; preserve each one's meaning.

Model-assisted finding, human verification

Van Bodegom says he supplied a suspicion about filter injection to GPT-5.6 Sol, which found the issue. He checked and verified the result. He also says Claude wrote much of the demonstration code, while he wrote the report's prose himself.

Those are useful details about the author's workflow. They provide no independent benchmark of autonomous security research. I would keep the verified report, the maintainer's acknowledgment and the installable fixes as the receipts that matter to the user. A debate about model rankings can easily crowd out the concrete permission check the repair added.

Verdict: SHIP IT

I'd ship the patched picker because it checks authorization at the control boundary, and I'd keep the current stable extension installed. Future security work stays open. The early update advice belongs alongside the explanation, rather than being held until the end of a security story.

For other privileged extension interfaces, I would ask who is allowed to command the interface and change persistent state. Reachability is one property to inspect. The caller's authority is another property that needs its own explicit answer.

FAQ

Which uBlock Origin version should I check? The advisory's patched baseline is 1.75.4, while GitHub and Mozilla currently list stable 1.76.4. Check the version actually installed through the browser's normal extension controls and use the current stable release.

Was the issue a zero-click attack on every browser? The documented path required a click and a relevant resource-serving filter in Firefox MV2. The report discusses Lite separately. We performed no exploit test.

What changed after the first hardening? The November 3 repair required a verified secret token for picker launch. Later changes added proposed-filter validation and resource-related restrictions.

Did an AI discover and exploit this without a person? The author says a model found the issue after he supplied a hypothesis, and he verified it. That account supplies no independent autonomous-research benchmark. The hidden-overlay click trap was described but not built by him.

Sources

The video's short silent caricature quotation comes from GIPHY, linked to a timbenj/Photobucket source. It references Nedry/Wayne Knight/Jurassic Park; exact animation authorship and original film-native footage are unverified. Underlying owners retain copyright. Hosting grants no reuse licence or blanket permission. The source report's displayed CC BY-SA4 licence applies to that article, without licensing the other source media or the whole episode.

This article expands on an episode of **The Daily Diff, a five-minute daily video on what shipped and what broke in tech.
Watch the episode · Subscribe on YouTube · the written diff lands in your inbox every morning at thedailydiff.dev.

Video transcript and primary sources